Fix symbol trampolines for OSX and export jl_n_threads

cli/Makefile

@@ -54,7 +54,11 @@ $(BUILDDIR)/loader_exe.o : $(SRCDIR)/loader_exe.c $(HEADERS)
 $(BUILDDIR)/loader_exe.dbg.obj : $(SRCDIR)/loader_exe.c $(HEADERS)
 	@$(call PRINT_CC, $(CC) $(DEBUGFLAGS) $(LOADER_CFLAGS) -c $< -o $@)
 $(BUILDDIR)/loader_trampolines.o : $(SRCDIR)/trampolines/trampolines_$(ARCH).S
-	@$(call PRINT_CC, $(CC) $(DEBUGFLAGS) $(LOADER_CFLAGS) $< -c -o $@)
+	@$(call PRINT_CC, $(CC) $(SHIPFLAGS) $(LOADER_CFLAGS) $< -c -o $@)
+
+# Debugging target to help us see what kind of code is being generated for our trampolines
+dump-trampolines: $(SRCDIR)/trampolines/trampolines_$(ARCH).S
+	$(CC) $(SHIPFLAGS) $(LOADER_CFLAGS) $< -S | sed -E 's/ ((%%)|;) /\n/g' | sed -E 's/.global/\n.global/g'
 
 DIRS = $(build_bindir) $(build_libdir)
 $(DIRS):

cli/jl_exports.h

@@ -5,9 +5,14 @@
 #include "../src/jl_exported_data.inc"
 #include "../src/jl_exported_funcs.inc"
 
-// Define data symbols as `const void * $(name);`
+// Define pointer data as `const void * $(name);`
 #define XX(name)    JL_DLLEXPORT const void * name;
-JL_EXPORTED_DATA(XX)
+JL_EXPORTED_DATA_POINTERS(XX)
+#undef XX
+
+// Define symbol data as `$type) $(name);`
+#define XX(name, type)    JL_DLLEXPORT type name;
+JL_EXPORTED_DATA_SYMBOLS(XX)
 #undef XX
 
 // Define holder locations for function addresses as `const void * $(name)_addr`

cli/trampolines/trampolines_aarch64.S

@@ -1,12 +1,27 @@
 #include "../../src/jl_exported_funcs.inc"
 
+// On macOS, we need to prepend underscores on symbols
+#if defined(__APPLE__) && defined(__MACH__)
+#define CNAME(x)    _##x
+#define PAGE(x)     x##@PAGE
+#define PAGEOFF(x)  x##@PAGEOFF
+#define SEP         %%
+#else
+#define CNAME(x)    x
+#define PAGE(x)     x
+#define PAGEOFF(x)  #:lo12:##x
+#define SEP         ;
+#endif
+
 #define XX(name) \
-.global name; \
-.cfi_startproc; \
-name##:; \
-    adrp x0, name##_addr; \
-    ldr x0, [x0, #:lo12:name##_addr]; \
-    br x0; \
-.cfi_endproc; \
+.global CNAME(name) SEP \
+.cfi_startproc SEP \
+.p2align    2 SEP \
+CNAME(name)##: SEP \
+    adrp x16, PAGE(CNAME(name##_addr)) SEP \
+    ldr x16, [x16, PAGEOFF(CNAME(name##_addr))] SEP \
+    br x16 SEP \
+.cfi_endproc SEP \
+
 JL_EXPORTED_FUNCS(XX)
 #undef XX

cli/trampolines/trampolines_i686.S

@@ -20,9 +20,9 @@
 
 #define XX(name) \
 DEBUGINFO(name); \
-.global name; \
+.global CNAME(name); \
 .cfi_startproc; \
-name##:; \
+CNAME(name)##:; \
     CET_START(); \
     jmpl *(CNAME(name##_addr)); \
     ud2; \

cli/trampolines/trampolines_x86_64.S

@@ -34,10 +34,10 @@
 
 #define XX(name) \
 DEBUGINFO(name); \
-.global name; \
+.global CNAME(name); \
 .cfi_startproc; \
 SEH_START1(name); \
-name##:; \
+CNAME(name)##:; \
 SEH_START2(); \
     CET_START(); \
     mov CNAME(name##_addr)(%rip),%r11; \

src/ccall.cpp

@@ -22,11 +22,11 @@ static bool runtime_sym_gvs(jl_codegen_params_t &emission_context, const char *f
     GlobalVariable *libptrgv;
     jl_codegen_params_t::SymMapGV *symMap;
 #ifdef _OS_WINDOWS_
-    if ((intptr_t)f_lib == 1) {
+    if ((intptr_t)f_lib == (intptr_t)JL_EXE_LIBNAME) {
         libptrgv = prepare_global_in(M, jlexe_var);
         symMap = &emission_context.symMapExe;
     }
-    else if ((intptr_t)f_lib == 2) {
+    else if ((intptr_t)f_lib == (intptr_t)JL_LIBJULIA_INTERNAL_DL_LIBNAME) {
         libptrgv = prepare_global_in(M, jldll_var);
         symMap = &emission_context.symMapDl;
     }
@@ -1266,7 +1266,7 @@ static jl_cgval_t emit_ccall(jl_codectx_t &ctx, jl_value_t **args, size_t nargs)
     auto _is_libjulia_func = [&] (uintptr_t ptr, const char *name) {
         if ((uintptr_t)fptr == ptr)
             return true;
-        return (!f_lib || f_lib == JL_DL_LIBNAME) && f_name && !strcmp(f_name, name);
+        return (!f_lib || f_lib == JL_LIBJULIA_INTERNAL_DL_LIBNAME) && f_name && !strcmp(f_name, name);
     };
 #define is_libjulia_func(name) _is_libjulia_func((uintptr_t)&(name), #name)
 

src/codegen.cpp

@@ -397,7 +397,7 @@ static const auto jlexe_var = new JuliaVariable{
     [](LLVMContext &C) { return T_pint8; },
 };
 static const auto jldll_var = new JuliaVariable{
-    "jl_dl_handle",
+    "jl_libjulia_internal_handle",
     true,
     [](LLVMContext &C) { return T_pint8; },
 };
@@ -7531,7 +7531,7 @@ static void init_jit_functions(void)
     add_named_global(jlRTLD_DEFAULT_var, &jl_RTLD_DEFAULT_handle);
 #ifdef _OS_WINDOWS_
     add_named_global(jlexe_var, &jl_exe_handle);
-    add_named_global(jldll_var, &jl_dl_handle);
+    add_named_global(jldll_var, &jl_libjulia_internal_handle);
 #endif
     global_jlvalue_to_llvm(new JuliaVariable{"jl_true", true, get_pjlvalue}, &jl_true);
     global_jlvalue_to_llvm(new JuliaVariable{"jl_false", true, get_pjlvalue}, &jl_false);

src/dlload.c

@@ -308,8 +308,10 @@ const char *jl_dlfind_win32(const char *f_name)
     void * dummy;
     if (jl_dlsym(jl_exe_handle, f_name, &dummy, 0))
         return JL_EXE_LIBNAME;
-    if (jl_dlsym(jl_dl_handle, f_name, &dummy, 0))
-        return JL_DL_LIBNAME;
+    if (jl_dlsym(jl_libjulia_internal_handle, f_name, &dummy, 0))
+        return JL_LIBJULIA_INTERNAL_DL_LIBNAME;
+    if (jl_dlsym(jl_libjulia_handle, f_name, &dummy, 0))
+        return JL_LIBJULIA_DL_LIBNAME;
     if (jl_dlsym(jl_kernel32_handle, f_name, &dummy, 0))
         return "kernel32";
     if (jl_dlsym(jl_ntdll_handle, f_name, &dummy, 0))
@@ -334,8 +336,8 @@ const char *jl_dlfind_win32(const char *f_name)
     // explicit is preferred over implicit
     return NULL;
     // oops, we didn't find it. NULL defaults to searching jl_RTLD_DEFAULT_handle,
-    // which defaults to jl_dl_handle, where we won't find it, and will throw the
-    // appropriate error.
+    // which defaults to jl_libjulia_internal_handle, where we won't find it, and
+    // will throw the appropriate error.
 }
 #endif
 

src/init.c

@@ -293,7 +293,8 @@ JL_DLLEXPORT void jl_atexit_hook(int exitcode)
 
 static void post_boot_hooks(void);
 
-JL_DLLEXPORT void *jl_dl_handle;
+JL_DLLEXPORT void *jl_libjulia_internal_handle;
+JL_DLLEXPORT void *jl_libjulia_handle;
 void *jl_RTLD_DEFAULT_handle;
 JL_DLLEXPORT void *jl_exe_handle;
 #ifdef _OS_WINDOWS_
@@ -657,7 +658,10 @@ void _julia_init(JL_IMAGE_SEARCH rel)
     jl_prep_sanitizers();
     void *stack_lo, *stack_hi;
     jl_init_stack_limits(1, &stack_lo, &stack_hi);
-    jl_dl_handle = jl_load_dynamic_library(NULL, JL_RTLD_DEFAULT, 1);
+
+    // Load libjulia-internal (which contains this function), and libjulia, explicitly.
+    jl_libjulia_internal_handle = jl_load_dynamic_library(NULL, JL_RTLD_DEFAULT, 1);
+    jl_libjulia_handle = jl_load_dynamic_library(JL_LIBJULIA_DL_LIBNAME, JL_RTLD_DEFAULT, 1);
 #ifdef _OS_WINDOWS_
     jl_ntdll_handle = jl_dlopen("ntdll.dll", 0); // bypass julia's pathchecking for system dlls
     jl_kernel32_handle = jl_dlopen("kernel32.dll", 0);

src/jl_exported_data.inc

@@ -1,4 +1,5 @@
-#define JL_EXPORTED_DATA(XX) \
+// Pointers that are exposed through the public libjulia
+#define JL_EXPORTED_DATA_POINTERS(XX) \
     XX(jl_abstractarray_type) \
     XX(jl_abstractslot_type) \
     XX(jl_abstractstring_type) \
@@ -115,3 +116,8 @@
     XX(jl_void_type) \
     XX(jl_voidpointer_type) \
     XX(jl_weakref_type)
+
+
+// Data symbols that are defined inside the public libjulia
+#define JL_EXPORTED_DATA_SYMBOLS(XX) \
+    XX(jl_n_threads, int)

src/julia.h

@@ -1515,6 +1515,7 @@ JL_DLLEXPORT int jl_is_debugbuild(void) JL_NOTSAFEPOINT;
 JL_DLLEXPORT jl_sym_t *jl_get_UNAME(void) JL_NOTSAFEPOINT;
 JL_DLLEXPORT jl_sym_t *jl_get_ARCH(void) JL_NOTSAFEPOINT;
 JL_DLLEXPORT jl_value_t *jl_get_libllvm(void) JL_NOTSAFEPOINT;
+extern JL_DLLIMPORT int jl_n_threads;
 
 // environment entries
 JL_DLLEXPORT jl_value_t *jl_environ(int i);

src/julia_internal.h

@@ -990,7 +990,8 @@ STATIC_INLINE uint64_t cong(uint64_t max, uint64_t unbias, uint64_t *seed)
 }
 
 // libuv stuff:
-JL_DLLEXPORT extern void *jl_dl_handle;
+JL_DLLEXPORT extern void *jl_libjulia_handle;
+JL_DLLEXPORT extern void *jl_libjulia_internal_handle;
 JL_DLLEXPORT extern void *jl_RTLD_DEFAULT_handle;
 #if defined(_OS_WINDOWS_)
 JL_DLLEXPORT extern void *jl_exe_handle;
@@ -1012,8 +1013,13 @@ JL_DLLEXPORT jl_value_t *jl_get_cfunction_trampoline(
 
 
 // Windows only
-#define JL_EXE_LIBNAME ((const char*)1)
-#define JL_DL_LIBNAME ((const char*)2)
+#define JL_EXE_LIBNAME                  ((const char*)1)
+#define JL_LIBJULIA_INTERNAL_DL_LIBNAME ((const char*)2)
+#if defined(JL_DEBUG_BUILD)
+#define JL_LIBJULIA_DL_LIBNAME          "libjulia-debug"
+#else
+#define JL_LIBJULIA_DL_LIBNAME          "libjulia"
+#endif
 const char *jl_dlfind_win32(const char *name);
 
 // libuv wrappers:

src/runtime_ccall.cpp

@@ -33,8 +33,10 @@ void *jl_get_library_(const char *f_lib, int throw_err) JL_NOTSAFEPOINT
 #ifdef _OS_WINDOWS_
     if (f_lib == JL_EXE_LIBNAME)
         return jl_exe_handle;
-    if (f_lib == JL_DL_LIBNAME)
-        return jl_dl_handle;
+    if (f_lib == JL_LIBJULIA_INTERNAL_DL_LIBNAME)
+        return jl_libjulia_internal_handle;
+    if (f_lib == JL_LIBJULIA_DL_LIBNAME)
+        return jl_libjulia_handle;
 #endif
     if (f_lib == NULL)
         return jl_RTLD_DEFAULT_handle;

src/threading.c

@@ -221,7 +221,6 @@ jl_get_ptls_states_func jl_get_ptls_states_getter(void)
 }
 #endif
 
-JL_DLLEXPORT int jl_n_threads;
 jl_ptls_t *jl_all_tls_states JL_GLOBALLY_ROOTED;
 
 // return calling thread's ID

src/threading.h

@@ -13,7 +13,6 @@ extern "C" {
 #define PROFILE_JL_THREADING            0
 
 extern jl_ptls_t *jl_all_tls_states JL_GLOBALLY_ROOTED; /* thread local storage */
-extern JL_DLLEXPORT int jl_n_threads;   /* # threads we're actually using */
 
 typedef struct _jl_threadarg_t {
     int16_t tid;